Next Buffett a bot?? May be… MIT model beats Wall Street analysts in sales forecasts

The curse of emotions

Warren Buffett once said, “If you have more than 120 or 130 I.Q. points, you can afford to give the rest away. You don't need extraordinary intelligence to succeed as an investor.” Temperament or emotional stability is more important, he meant. The best part about a machine-- its emotional neutrality, no biases no favorites.

 Various people tried creating models to beat his returns. Thousands of books were written on various stock selection models, but none succeeded, as continuous learning is essential to learn the novelties of stock markets. But, as machine learning and alternative data develop at sky rocketing speeds, it has become impossible for human brain to process huge amounts of data. A computer on the other hand has no limits.

 MIT has recently developed a model for predicting sales using NOISY data, and it outperformed wall street analysts significantly.

But wait, Noisy data…what could it mean?? Let’s try & understand

The relevance of noise

Put yourself in fund managers’ shoes for a sec - say you wish to invest in a mall chain. To buy at the right time, you need to estimate the next quarter’s figures. Is there a way to know how well it will perform today itself?

 Turns out there is:- A satellite image of the mall’s parking vs competitors mall parking can say a lot about the footfall. What if you also have a footfall to revenue ratio for the mall or for the industry in general? You could make a fairly good estimate about the sales too. And, just by using the so called “noise”.

 Is this imaginary project? Nope..companies have started doing this, financial managers are spending billions to purchase data, and numbers are just going to increase. This noise is what we call ‘Alternative Data’ – basically anything and everything not directly connected to the outcome, but can be used to make an educated guess-like footfall for sales.  

 Have a look at a few headlines to understand how quickly its growing:-

More industries to Adopt Alternative Web Data in 2020

How Big Investors Cash In on ‘Alternative Data’

HEDGE FUNDS SEE HUGE POTENTIAL IN ALTERNATIVE DATA

Wall Street is chasing a data gold rush. Here's our deep dive on its efforts to crack the code.

Investor spending on financial data jumps to post-crisis record

 MIT Model

In a paper published at the, researchers published a model which estimated quarterly earnings. The task was simple- predict sales numbers of 30 companies for the coming quarter.

The analysts had unlimited access to any public/private data or machine learning algorithms. While, the model just had a tiny data set of 2 types.

---Anonymized weekly credit card transactions

--Quarterly reports

 Result- the model outperformed the Wall Street analysts on 57 percent of predictions.

Great but….how does the model work?

Agreed the model performed fairly well, but how can sales be determined by credit card sales numbers?

 --Don’t people use debit card/cash/digital wallets for shopping?

--If you say credit card sales account for 10% of sales for a week, how do you know its 10%? Since, you can't access weekly sales figures.

 The model has a really complex math, known as Kalman filtering. Simply put, the model breaks previous quarters sales into 90 days, allowing the sales to vary on day-to-day basis. Then, the credit card transactions are matched to sales. Using extrapolation the model finds a relation between credit card transactions and sales. Then, it calculates each day’s fraction of observed sales, noise level, and an error estimate for how well it made its predictions.

 All these calculations are made on daily numbers. Once, the data points are ready they are finally put into the inference algorithm to estimate the daily sales. Not to mention, you simply multiply the number of days to get weekly or quarterly figures.

 Honestly, I couldn’t understand the entire calculation fully. But, the bottom line, it works well for not one or two companies, but on 30 companies, beating Wall Street analysts over 50% of the times.  

 Want to dig deeper here’s the link

 https://scitechdaily.com/automated-system-beats-wall-street-analysts-in-forecasting-business-financials/

 Are machines better financial managers?

 Buffett once said--Most investors including financial managers fail, not due to knowledge, but because of emotions, not due to IQ but because of EQ. Given, the superiority of machines in terms of IQ, and advantage of having no emotional issues, can they take over most functions of fund management in future?

Either time can answer…

     OR

You can, through the comments section…

 Goodbye…keep wondering